LF4CV@CVPR2017

The current program is work in progress and subject to change

8:50: Welcome Address

Session 1: Invited Talks and Poster Session

09:00: Keynote: Ravi Ramamoorthi (UC San Diego, USA); Light Fields - From Shape Recovery to Sparse Reconstruction; The availability of commercial light field cameras has spurred significant research into the use of light fields and multi-view imagery in computer vision. In this talk, we discuss our results over the past few years, focusing on a few themes. First, we describe our work on a unified formulation of shape from light field cameras, combining cues such as defocus, correspondence, and shading. Then, we go beyond photoconsistency, addressing non-Lambertian objects, occlusions, and material recognition. We also discuss applications for light field cameras such as motion deblurring and descattering. Finally, we show that advances in machine learning can be used to interpolate light fields from very sparse angular samples, in the limit a single 2D image, and create light field videos from sparse temporal samples.
09:45: Keynote: Thomas Nonn (Lytro, USA); Light field for Cinema; Lytro has developed a professional light field solution for cinema, encompassing capture, processing, visual effects and data storage. Light field cinema expands on the current tools available to cinematographers by providing unsurpassed post-processing flexibility and unprecedented control over focus, aperture, perspective and shutter angle.
The presentation will begin with an introduction of the Cinema camera, outlining some of the design decisions that were made to ensure high image quality and maximum flexibility. Lytro's light field cinema camera generates a very rich dataset which places storage and processing challenges on both local networks and the cloud. Processing this rich dataset requires an optimum balance between performance and quality. Visual effects tools such as optical flow, camera tracking, segmentation, and point clouds are but a few of the operations benefiting from light field.
10:30: Poster Session
12:00: Lunch Break

Session 2: Invited Talk and Paper Presentations

13:30: Keynote: Harlyn Baker (Electronic Visualization Laboratory at University of Illinois Chicago); EPI Imaging on Contours for Range with Structure; Epipolar-Plane Image Analysis was developed years ago as a means for acquiring accurate and precise scene range information from a camera moving along a linear path. EPI was a precursor to digital light-field analysis, showing the benefits of constrained redundant imaging. At the time we developed from a structured image-analysis perspective, forming the EPI volume and using linear features as our estimation observations, extracted from a connected manifold. Cameras then were expensive, large, and bandwidth challenging. Advances of mobile computing have changed all this with tiny free (essentially) cameras and speedy processors and busses. While the developing light-field community has taken a retinotopic perspective in their analysis - filtering on EPIs for depth - the original approach retains advantage. Principal among these are the retention of spatial continuity in the representations, a focus on discontinuities, and the ability to infer and model anomalous events such as specularities and occlusions. I will discuss some of the roots of EPI, and our more recent efforts at bringing it back into operation with modern tools.
14:15: Katherine Skinner (University of Michigan); Underwater Image Dehazing with a Light Field Camera
14:35: Donald Dansereau (Stanford University); Richardson-Lucy Deblurring for Moving Light Field Cameras
14:55: Gregoire Nieto (INRIA, France); Linearizing the Plenoptic Space
15:15: Coffee Break

Session 3:Benchmark, Challenge, and Panel Discussion

15:45: Keynote: Tsung-Yi Lin (Common Visual Data Foundation); Common Visual Data Foundation: Enabling Community-driven Research in Computer Vision; The Common Visual Data Foundation (CVDF) was established in 2016 with a mission to enable community-driven research in computer vision through creation of academic datasets and corresponding competitions. In this talk, we will first share our experience in building Common Object in Context (COCO) dataset with focus on collecting quality annotations, designing evaluation metrics, and holding competitions. We next discuss the recent activities and plans of CVDF for helping our community to create and host visual datasets.
16:10: Katrin Honauer (Heidelberg University),
Ole Johannsen (University of Konstanz); New benchmark: challenge evaluation and state-of-the-art; In this talk we will present the current state-of-the-art of light field depth estimation based on the challenge submissions and raise questions about the advantages of light fields.
16:30: Talk by Outstanding Challenge Participant I, OBER, Hendrik Schilling
16:40: Talk by Outstanding Challenge Participant II, PSRF, HaeGon Jeon
16:50: Talk by Outstanding Challenge Participant III, RM3DE, Marco Carli
17:00: Talk by Outstanding Challenge Participant IV,SPO, Shuo Zhang
17:10: Short Break (to set up discussion)
17:15: Panel Discussion; This discussion will cover the following topics: benchmarking, goals of light field research, datasets
18:00: Closing Remarks